Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Again about some details of Kasparov vs. DB in 1997

Author: Hans Gerber

Date: 19:18:15 06/22/00

Go up one level in this thread


>Why?  They knew they were not cheating, and all computers play very well in some
>positions and like amateurs in others!  Look at the recent Fritz games where
>Fritz could not see the danger to his king, which Kramnik saw easily.  There is
>nothing unusual in this sort of spotty play.  The fact that DB avoided Qb6 in
>game 2 is nothing special either, given that it has some 6000 evaluation terms
>and is a totally different chess entity than other programs.  Kasparov had no
>*logical* grounds to suspect cheating.

Are you sure? Could you prove that? How do you know that Kasparov could not gain
certain insight into the machine's thought process by e.g. the time it used for
each move?

Of course the mere numbers have impressing results. But are you sure that it is
allowed to assume a qualitative difference and not just a quantitative? The
games itself IMO do not support your assumption of a totally different entity
with a totally different play. In that context it is telling that they did not
publish the many training games they played befor the match.

>No it isn't.  This was not a scientific experiment.  There was a general ACM
>Computing Challenge goal to see whether a machine could beat the human chess
>champion at tournament time control, but it was not a controlled scientific
>experiment and was never billed as such.  It was a publicity event, held for
>money.

Agreed it was not a controlled scientific experiment. But let me declare that
even in a publicity event the scientists of the DB team are not allowed to
violate the minimum of scientifical standards. Also in a publicity event the
rules of logic are not deactivated.

Let me repeat that if you want to see what your machine could do against a
strong chessplayer, it is unfavorable if you as the team start a psychological
fight with the chessplayer. Is it so little known what already the duties are in
view of a guest? Not to speak of a real favorite who Kasparov was for a long
time. It is simply unbelievable how mean they treated Kasparov. Let us wait
until we know all the details of their mistreating Kasparov.

Since you are very careful with your conclusions about the final result of the
show and its relevance why can't you agree that especially the topic of the
logfiles destroyed the meaning of the outcome. Always remembering that a
chessplayer without being motivated is a bad candidate for a test. Are you sure
that the ACM has meant such a thing?

>Come on, you are imposing conditions that simply did not exist.  It was not a
>science experiment.  There is no way to control for the emotions of the human
>player in this setup.  Kasparov openly admitted that he was afraid of the
>machine, for good reason.

But not the ones you seem to suppose. Not because of the objectively existing
strength. However because of the secrecy before the match. The amount of
advantage through the secrecy alone could not be exaggerated! It has nothing to
do with normal chess in today's tournaments. Remember that GM Benjamin tuned the
machine against Kasparov many months long. Spooky.

Of course with no general scientifical value. Or you should call for 15 versions
of DB if it would be allowed to participate in a modern tournament. Tuned
against 15 opponents. That makes 7 and one half years...  Excuse me I'll be
right back again...   :-)


>The gold standard of a scientific experiment is the
>double-blind, and there was not even the semblance of any sort of controlled
>experiment in this match.  Perhaps you could set up a match among top GMs, yet
>not tell any of the players who they will be playing against.  You also do not
>tell them that there may be computers involved.  The conditions must be set so
>that no one knows who is playing whom in any given game.  This is probably the
>closest you could come to determining how well a machine can really play
>technically against humans.

You are making a big mistake! Permit me now my own suspicion. Such a tabloid
could only be invented by someone deeply involved in computerchess. Very telling
indeed.

I must remind you of the fact that the double-blind would also require that your
machine had to play on its own! Its programmer not being informed where it was
playing...

But honestly I can not understand you.

They wanted to know what the machine could do. They wanted to try it against
Kasparov. For reasons of his strength and attraction. The games of Kasparov are
well known in the world of chess. Now the only thing the DB team should have
guaranteed is that Kasparov could have played in his role as favorite guest and
star of the show.  The moment he felt being treated as sort of enemy Kasparov
realised that he was not prepared for such a fight. He explained all this in his
many speeches.

Now we had several here in computerchess who argued that then Kasparov should
have been better prepared, for all not relying too much on Fritz and Hiarcs...

Such arguments have no meaning because you can't prepare on opponents you have
never seen before. Nonsense, the people replied, there are enough games of prior
versions of DB. On the other side it is you who showed the importance of "6000
evaluation terms and a totally different entity". Now you must decide. Either it
is true that no preparation was possible or the older games could have been of
great help...

> Obviously the DB match was nothing close to this,
>and it was not billed as such.  Of course all the psychological stress would be
>on Kasparov alone. So what?  If he were allowed to play the same machine as
>often as he liked, and the tuning on the machine was not changed, eventually he
>would determine the weaknesses of the machine and prevail.  So what?  I should
>think most chess people believe that Kasparov is actually stronger than the
>machine, but the stronger player doesn't always win a match.

Which had to be proven in case of Kasparov. Since he has won all the matches or
rematches at least.

> It certainly
>didn't prove to me that machines are stronger, it gave an example of what I
>already believe, i.e. that with sufficient power a computer has chances to win
>even at the highest levels, and these chances will continue to increase with
>computing power.  Big deal.  Who doesn't know this?  DB put to rest *most* of
>the doubters that said many years ago that a computer would never even come
>close.

Please don't hurry through history! For me it is still a question how even
non-GMs would dance with DB if they had the time to prepare with the machine.
For me it is still not proven by the little portion of Kasparov games that DB
had no major weaknesses. I am not sure if the respect for the machine is
justified and that a in-depth-analysis could not reveil typical computer-like
weaknesses.

Perhaps this is the reason for the general secrecy!


>Computerchess will continue with or without Kasparov.  If one day someone builds
>a supermachine successor to DB that plays so well that it needs no further
>tuning, then by all means you can put in on the Internet and have it play the
>whole world 24/7, and people can study its games, etc.  A six game match does
>not prove that computers have won, and as I said I think most people agree human
>adaptability will produce superior results if the machines still have blindspots
>as they still do (witness Fritz's losses the other day).


Finally we have agreement again.  :-)


> I think right now the
>game in computerchess is to be able to beat the best humans consistently at long
>time controls on relatively "normal" hardware.  I think even Kasparov would
>admit that advances in computerchess will continue even as humans largely stand
>still in their chess skills.  Kasparov vs. DB2 was not the end, so I don't think
>Kasparov should have gotten so excited about it anyway.  He just really really
>hates to lose. :)


Very true. But you are unfair if you don't accept that Kasparov had his
experience in the second game at a time when he was not losing but leading by 1
- 0. Note that Kasparov had played a game where all his GM collegues would have
ripped him into parts. That alone shows that Kasparov tried to find out what the
machine could do. He did a great job as a tester. But then they threw him out of
the boat.


>>Not that I wanted to say that you really needed such extra lessons, but we talk
>>in front of many spectators...
>
>Maybe, but I think any spectators here have made up their minds on Kasparov vs.
>DB years ago.

Therefore it's a demanding task to show the importance of the collection of data
you build your judgement on. Prejudices should not be confused with mature
conclusions. Not that I wanted to say...


Hans Gerber

P.S. Please read the text of the press conference after game three. Read also
what M. Ashley insisted to ask Kasparov and look what Kasparov replied. It is
the best document to show that Kasparov did not do or imply what you have
written. In connection with the informations we got about the multiple
possibilities where a cheating could have happened, you must accept that
Kasparov did not accuse the DB team. This is only the case if you reduce the
possibilities on the "war room". Please give me in the first instance your
reflections on that data. I want to be sure that I did not miss something
important out of language reasons again...




This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.